jules-general・🦑

[cyan bronze]

I've had that one once in the past but after arguing with him I ended up managing to have him do it, don't remember how tbh and I cleaned old chats like an hour ago...

[rough tendon]

pushing the changes ends the task?

[cyan bronze]

nope and yes

[rough tendon]

hahaha

[cyan bronze]

it's marked as complete be you can still work on the task

[rough tendon]

okay okay lemme try

[cyan bronze]

there's no real limit on what you can do on a single task I sometimes do multiple things in a single chat but that ussually causes allucinations

well there's a message limit per tasks actually

but it's high

[rough tendon]

yea thats the problem i think i gave it like 5 tasks in 1 task ;-;

[cyan bronze]

sometimes it works flawlessly you know...

but it's "sometimes"

Jules is frustrating because you are left with debugging but it's very usefull

[rough tendon]

i think i found a wrok around

not so good but works

i asked it to type the whole updated code in chat

lmao

it gave me the updated code

[cyan bronze]

that's a mad one. never heard anyone do that one

But I'll remember it just in case

[rough tendon]

its like torturing the bot to type the whole code in teh chat

hahaha

[cyan bronze]

[cyan bronze]

my average experiance using Jules is the same as bullying the intern on his first day

[rough tendon]

hahaha

great my codes working now too

jules op asf

[cyan bronze]

yeah he works well on large tasks

small tasks and css tho...

better not

[rough tendon]

fr he is doing pretty good anyways this bug testing thing imma make him do 5-6 task in 1 task

lmao

[rough tendon]

thanks for helping tho

[cyan bronze]

[formal tusk]

hmmm please elaborate on how Jules is "arguing"

[cyan bronze]

well when he's hallucinating he submitted changes he tends to be stubborn and say he did. Sometimes I tell him there's a fault in your logic, explain the fault exactly using the variables and file names to elaborate and he says it's irrelevant to the current issue because he thinks the issue is caused by something else...

there's a few situation when it's just hard to make him do what you ask

I deleted a few chats today sadly or I would have provided screens but it's not possible at the moment

if I run into an argument with him again I'll make a thread in the feedback section.

yes also sometimes he just wants you to test something and won't let you proceed unless you do. he'll just ask again saying it's a crucial info when really it's not

my point being he's just really stuborn

ah and I missed the forest... there's that one recurring issue where he wants the user to do the test before he provides the code and give some "sorry I'm late my alarm didn't ring" kind of excuse when you tell him you can't do that because it's just not possible.

[quasi finch]

@formal tusk are you the person we make suggestions for features/improvements to?

[rain phoenix]

You can give Jules something "broad and shallow" or "narrow and deep" and it should be okay, but cramming a bunch of tasks into one is gonna be difficult

This might prove to be a little bit "too broad"... we will see!

[cyan bronze]

if that makes you feel better I gave him that stupid one an hour ago can't wait for everything to break lol

[rain phoenix]

https://tenor.com/view/la-noire-valdez-are-you-a-madman-madman-gif-13941016901046598736

[cyan bronze]

https://tenor.com/view/god-no-stop-chemistry-danger-gif-8954456262652807740

hold on let me cook

[rain phoenix]

https://tenor.com/view/van-crash-car-crash-sfm-gif-20920412

[cyan bronze]

burning intensify

[rain phoenix]

someone above asked about adding mcp support

you can just give jules access to a command line mcp client

you could even have jules code one for you

the vm itself kind of obviates the need for a lot of mcp however

update: jules nailed the docblock request, i feel like i finally understand my own ai slop code

[cyan bronze]

surprise mine failed the request.

[rain phoenix]

I expect these docblocks to really help with future vibe coding also

along with vigorous test suite, AI ever expanding

AI is getting me to 100% code coverage

[cyan bronze]

wait... I made the same request I just removed the "if" part at the end and now he's doing it. Jules is a madman

[scenic coral]

Is there an API to call Jules? Can I run Jules tasks from CLI?

[burnt heron]

anyone got any good jules prompts?

[delicate fog]

Definitely needs a "stop thinking" button, or even better, like cursor, just let people send messages even while the agent is thinking 🤦

[worn jasper]

(empty repo) i gave permissions still im not getting any repositories what i can do ?

[spare merlin]

after a while you can send a message again, and it will interupt it's thinking and add your message to the progress

[vapid moat]

What I do is get natural language, feed it into Gemini 2.5 Pro and ask it to break it down into a dual task list because it seems to intuitively understand being another Google tool what sort of task list it actually finds best.

[vapid moat]

What, you mean like system prompts at first? I never thought of that.

eBay asked it to state it's an operating environment and set itself up to work with blah blah blah the languages you want the tool sits you need etc and ensure that everything is easily accessible all variables and are set path is set etc I don't know

I shall think on this in the three seconds of day I get.

[ebon oracle]

Can you post this in #1374125660859142254 please

[rapid pewter]

has anyone had luck with jules building a full front-end+back-end app? like airbnb clone with auth and sign in? doesn't work for me. what are the examples of the apps you have built that have front and backend?

[cyan bronze]

luck?

jules can do it, but it still takes R&D not luck

there currently isn't an option to build platforms by magic if that's what you look for.

If you can handle the debugging Jules can do what you said I think. you'd have to work on it full time for a few weeks/months and you'd still have to have to outsource the logo and most UI elements.

[sudden drum]

What's jules

[elfin kite]

learn more about Jules by checking out our latest announcement! #labs-announcements・📣 message

[sudden drum]

Ahhh

[humble latch]

Getting command timed out whenever it tries to run tests, and the whole vm becomes unresponsive, and then it can't even commit the changes. Today run in bash tool call internal errors. Maybe it's because overlayfs eventually runs out of space because it is copy on write even though df -h says 70 percent available.

Also the code quality is not that good compared to using Aider and 2.5 pro directly, because Jules doesn't give proper context to its workers

[quasi finch]

Jules will occasionally insist it can't push code to my repo and will give me code to copy and paste or will ask me to manually adjust code. Is this a known bug or should I create a new entry in the #1374125660859142254 channel?

[cyan bronze]

there's a fewq feedbacks about that

like one of them for visibility if you can

[cyan bronze]

but I was speaking about it there too

it's known i hope

[mint heart]

For the last few hours my task has been fluctuating between "booting up vm" and "setting up the repository". For the last few days I had to spawn numerous tasks as each task decides to quit after a while.

[formal tusk]

I'm one of them! You're welcome to also please put it through the jules-bugs-feedback channel as we pull everything from there to triage too.

[formal tusk]

Not yet!

[formal tusk]

Have you tried building with Stitch and then making updates/bug fixes with Jules? Could be cool.

[scenic coral]

Thanks for the reply. Is there a timeline for that?

[formal tusk]

We have a "jules awesome list" here: https://github.com/google-labs-code/jules-awesome-list

Feel free to contribute directly as well! We do want this to be community serving.

[scenic coral]

Also, what about the editor or code reading part of the story? This is actually the main reason I thought about the api.

[rapid pewter]

hm.. there is less control then in aider.chat or cursor. starting vm is a good feature.. just not sure where to use jules

[formal tusk]

Current known issue! We are investigating/addressing this.

[modest crater]

Hello

[cyan bronze]

[high holly]

Today got access to codex with Plus subscription! I immediately ran to test it it solved a problem that Jules couldn’t handle and I haven’t noticed any bugs so far hope that Jules will fix all his bugs soon and move on to improving Jules

[dry glade]

I do belive they should teach to IA about Ex Nihilo, Ad Perfectionem and make his biggest way to be better

[fast needle]

yeah I dunno. Jules has been unusable for me the last couple days 😦 Codex works fine even though no internet. Funny thing is this is completely the reverse from when I first started trying a few days ago 🤷

[high holly]

A familiar problem… the only thing that helps here is creating a new task 😄

He also loves to refuse quite often too)))

I remember one time when I asked him to do something and he said he had a fundamental error. He said the issue was caused by the cat command because he was trying to read a file but ended up reading a directory which caused the error. He even understood that he was using cat on a directory and couldn’t do anything about it. I asked him to stop and not act dumb and even pointed out the problem but he said he couldn’t do anything else because the cat command gets executed before he can intervene.

[fast needle]

yeah I think they are trying to run tasks/turns without a real vm most the time, causing these types of calls to fail. oh well, i'll keep trying every few days i guess

[unborn glade]

Hi

[wraith frost]

I am having a problem where Jules states it has run out of space for python packages.

[formal tusk]

May I ask what the task was? Curious to see (feel free to dm me the dets if that's preferable).

[muted rapids]

The daily task limit currently says "6 of 60" but the "+" button icon is disabled and has the helper text "You have reached your simultaneous task limit of 5 tasks". Is this a bug?

I guess asked another way: at what point is a task considered "In progress" vs "done"?

[cyan bronze]

https://discord.com/channels/1172568727942860810/1374125660859142254

[muted rapids]

looks like reloading the page worked 😄

[formal tusk]

Send screenshot? Alternatively send through the feedback button in the product(top right corner).

[past hill]

Jules is not smart at all... I feel like it is Gemini 2.0 or 1.5...Because my project is clearly using tailwind css but it generates this:

No offense... I cannot even use Jules even for a starter project because it is not useful.
I can get 10x or 100x better results using Gemini 2.5 Pro or Sonnet in Cursor/VSCode...

[wanton pecan]

Does Jules not work in India? I installed it on my repo and all I see is this. No way to select the repo I want to work with. Is this a known bug?

[iron mirage]

interesting. I just tried installing Next.js with Jules.
But it tries to install Page router. even tho I said the Nextjs latest version

and I just figured out it cannot install the next.js project on empty repo

amazing!

maybe I need to tell it has to use npx create-next-app@latest command

it seems like there's no way to delete repo that I added on Jules?

[iron mirage]

it seems like

npx create-next-app@latest .

installing Next.js project in the current directory does not work.
but this works.

npx create-next-app@latest test-app

if you give the name of the project then it works.

[runic valley]

Hey guys,
I've been experiencing something weird. The execution plan is presented to me and while I'm reviewing it, it gets accepted for some reason.I didn't touch the approve button. It happened to me on desktop and mobile multiple times already. Am I the only one?

[quasi finch]

I’ve noticed a countdown sometimes where a plan will get auto-approved in two minutes and counting

[iron mirage]

this happened to me one time today

[iron mirage]

Let's report this here #1374125660859142254

[runic valley]

Is there a timer?😮

[quasi finch]

I’ve noticed one twice. There’s been plenty of instances where there was no timer. Jules waited for me to approve the plan

[high holly]

My task was to modify the DDNet client. I wanted to test Jules’ capabilities in different types of tasks. Previously, I tested him on creating Minecraft mods, and he performed quite well in that area.

[cyan bronze]

there's a timer It's most likely intended behavior

pause the instance if you need more time

[high holly]

Did you check if you connected your GitHub account and granted the necessary permissions? Like access to all repositories? And did you give access to the correct account?

[limber raft]

Are there any known "best practices" in terms of writing prompts for Jules? The recommended prompts (from the Docs) are all very basic like "Add a loading spinner while fetchUserProfile runs". I tried those but not much success. I've since started to write more comprehensive prompts, similar to what I would do in Gemini and got a bit better results.

[brisk viper]

What I have found useful is to maintain a txt or markdown file of the exact things that need to be done: I create this using Gemini.
For instance, a MD file for the feature set and general outlines of what needs to be done. And, maybe a milestone-based doc to break down the development steps.
Then, upload this to the repository where you are working and ask Jules to refer to this documentation to work on milestone 1 or something like that.

Made my life a lot easier.

Additionally, in my first prompt, I ask it to maintain a log of what worked and what did not work and which milestones it achieved so that I can ask it to reiterate or fix it later on.

[limber raft]

Thanks, that's a great idea, I'll give that a try! I started structuring my prompts with the following headings like: Objective | Steps | Post-tasks | Prerequisites | Rules . It's starting to work better than those examples they gave in the docs but now my problem is Jules seems extremely extremely slow. But I'll give the MD file idea a try.

[brisk viper]

Awesome! I hope it works for you.

[muted rapids]

Yep I can provide feedback. I think the issue is similar to others I've seen reported, which is that the "+" button and helper text is not dynamically updated / synced

[formal tusk]

Hey! Can you confirm that the repo isn't empty?

[formal tusk]

there's a 2 minute timer and then auto-approves.

[formal tusk]

Yes yes -- it's 2 minutes! We heard that the 30 seconds one previously was just bananas so we updated it to 2 minutes.

[formal tusk]

oh yes yes! This is a lot of the polish on our list. Thanks following up.

[kind moth]

Hey jules team

I've recently had no issues at all with Jules

It's been performing really well

I just used it on an actual hobby project today with almost no issues

There is a little annoyance tho which is avoidable, but might be a bug or not be a bug, I'll post it in feedback hopefully you can tell me if I'm doing something wrong
Edit: https://discord.com/channels/1172568727942860810/1379862156560040099

It's really amazing how many things it takes into consideration without me even needing to tell it, great job 🎊 🥳

[cyan bronze]

it was lol, in most cases the page wasn't even loaded yet. I bet most people waren't even aware they have the option to refuse the plan back then most of the time I have like 1,15min left after loading. tbh I just paus the instance till I'm done reading.

[worn jasper]

Same i got this but my friends got repos I didn’t got any repos. So its working in india but why this issue anybody

[rain crown]

Hi, when will Chrome support ENS domain names? Supporting Web3 would be a great step forward

[chilly spindle]

@wanton pecan Jules is available in India. Have you tried connecting your github account to jules. See https://jules.google/docs#connect-github @worn jasper

[rain phoenix]

would love if this were user-configurable. Even 2 minutes gives me an "under the gun" feeling lol

here's the Jules prompt I use for my repo.

# Task

We are working on code coverage in src/Message/JsonRpcMessage.php

Please address the following branch

fromJson
- isset($data['result']) || isset($data['error'])

## Background

You are working on an MCP (Model Context Protocol) Server in PHP.

## Getting Started

You will want to run jules_setup.sh once to get started and be sure that run_tests.sh passes before you start and passes when you are finished. However, in the middle of coding, it's perfectly okay to run individual tests instead of the entire shell script.

Before working on this code, after reading README.md and JULES.md, read the typescript schema for the MCP protocol here:

https://github.com/modelcontextprotocol/modelcontextprotocol/raw/refs/heads/main/schema/2025-03-26/schema.ts

The MCP Server must NEVER violate this schema.

## Requirements
- The first step of your plan MUST be to run jules_setup.sh
- Do not attempt to add any new tools or resources to src/
- Tools/Resources you need for the test suite should be put somewhere in tests/ - Keep them simple!
- Avoid writing to STDOUT or STDERR during unit testing, let's see a clean "field of dots"
- Use PHPstan ignores freely, do not sacrifice code quality to make this tool happy!
- No code will be accepted unless run_tests.sh passes!

It works pretty well and code quality is pretty high coming out of Jules with this prompt. Of course you have to tailor it to your specific codebase and its particulars

Getting it to run this script consistently instead of trying to do its own thing was a bit challenging but as long as it's an explicit step it will obey

(by design this script doesn't send a lot to STDOUT, my thinking is that if we thrash the context with useless output it might affect reliability later on, although I assume the architecture tries to account for this at least a little bit already)

If anyone is working on a PHP library and wants a copy of this script, it is available here: jules_setup.sh

I don't know if this is helpful for anyone else!!

but I hope so!!!

it would be nice if I could simply provide the content of step 1 of the initial plan rather than having jules do it

[rain phoenix]

I adjusted the Task header in my Jules prompt to have this specific guidance underneath for test coverage tasks

You may have trouble reaching these branches. If they are truly unreachable, they should be deleted. If they are only reachable by treating docblock param types as suggestions, do so and suppress the PHPstan warning in the test case. If they are only reachable under irreproducible conditions, add a comment to suppress code coverage.

now I just need a way to create Jules tasks programmatically 😈

I hope Jules understands the difference between situation 1 (unreachable) and situation 3 (irreproducibly reachable)

then again,

alright enough thought-dumping, i'll yield the floor

[formal tusk]

Good feedback! I remember when it was 30 seconds I could barely click into each drop down so pausing was definitely helpful.

[formal tusk]

only one way to find out - give it a spin and let us know!

[rain phoenix]

it was definitely struggling with these before I added that guidance.

I'm not sure how familiar you are with PHP but here's how Claude suggested to reach these branches

but elsewhere it said Branch 2 can happen on certain buggy versions of Microsoft IIS

If it didn't make static analysis freak out because a string is NOT a string | false I wouldn't even bother with these lol

[chilly spruce]

@formal tusk I've noticed Jules getting so sluggish it's not worth the time? With my project Jules cannot create directories in a branch of a downloaded repository and advises me that its executable environment is not functioning - is this normal? Too many users?

[formal tusk]

Hmm would you mind dming me and/or share the feedback for the task in the product feedback form? Hoping this isn't a bug I've been seeing today.

For sluggishness, I can take a look internally at our traffic control.

[ashen ice]

Jules 什么时候支持多模态输入“?比如 截图

[mint heart]

I have so many - Jules was unable to complete the task in time. Please review the work done so far and provide feedback for Jules to continue.

[vapid moat]

I suggest reviewing the work done and giving feedback and then hit it continue.

[vapid moat]

I can confirm it is exceedingly sluggish. I mean, I would expect this given it's such a complicated program, but we're talking response times with like 10 to 30 seconds.

[mint heart]

Thanks. I have been blindly publishing the branch and providing positive feedback to Jules to continue to next step. I am just trying to test if Jules can build an entire application based on vision documents. So far, this is where it has reached. I have not tested the code it has generated but I just keep asking Jules to plan and implement the next steps every time it stops.

https://github.com/psreek-ai/AIFirstPodcastApp/tree/jules_wip_15750240882993844320

[vapid moat]

Ph ni

Like i got windsurf,itll find 3-4 bugs each time

Gotta review

Oh it absolutely can build a full application buddy, but it does not build it book free.

Bug

This tool is amazing. I really am not looking forward to when they start charging you $100 for it. I'm going to negotiate and say, I don't know, hey, I'll keep putting nation state malware disassemblers through if I get free access. Come on, guys. Help me out. I'm poor, but I do good things. That's my plan. We'll see how it goes.

KP14 : UNPLUGGED I'm going to release soon as soon as I get authorization from the government. It's a disassembler that was proven successful at taking apart the APT-41 key plug and it should work on a variety of other things using a lot of machine learning technology.

I mean you've got to admit decompiling and reassembling into source code a piece of nation-state grade malware is a pretty impressive advert.

[iron mirage]

Please fix this issue. Duplicated repos and I deleted all of them but they are still there.

[keen flare]

Hello, does anyone else have this problem loading the tasks page?

[kind moth]

No

[fossil karma]

I had that a couple of times, it never wanted to start up, I created a new task on that branch.

[mint heart]

Happens all the time for me after a few iterations. The issue with starting a new task is that the context is lost on where the previous task was.

[mint heart]

I have also noticed that sometimes the commit messages mentions changes to files but the commit shows that the those files were not changed

[cyan bronze]

I think the funniest/most-annoying part is that Jules genuinely believe that his last changes addressed the issue.

[formal tusk]

很多人问过要多模态输入！肯定是个巨大的挑战。。。不过肯定在好好的考虑 😅

[formal tusk]

ok we're seeing this and we hear you -- team is currently sprinting to address the latency issue. Stay tuned!

[vapid moat]

I mean, I'm not bothered. It's frankly, considering what you get, not bad. It's just, I mean, if you're looking for perfection, that would be the thing that's doing it.

I'm fairly sure this viewpoint is shared.

[formal tusk]

thanks for sharing! Would you share this in the feedback button (top right) so we can dive into it. This looks like a funky bug.

[vapid moat]

That's why at the end of each task you have to put a concise summary of what you've done in that session and where you're at.

Then feed that right back in. The key part of especially this AI is planning, planning, planning.

It's an assisted, not a replacement.

[formal tusk]

https://tenor.com/view/mashle-dance-gif-9745779996362916675

I'm not going to openly say we're looking for perfection but......

[vapid moat]

Honestly, I'm shockingly impressed with how effective this tool is. I realise it's not always generally free, so yeah, you saw my plan above of bargaining to get it free because I feed through some pretty interesting data, I'll see how that goes. But yeah, this is really something quite interesting.

Could I just feed in a straight shell script? I'm guessing not, I'd have to put it into a repository. There's one particularly annoying one I need to fix.

[tame atlas]

When will we be getting a Jules agent for vscode? 👀 seems like googles code assist could really use a good update

[vapid moat]

A while

Look st how it wirks

Full vm per req

It isn't really designed as an agent in that way. It's designed to be given a task and then to get on with it. Not like a question by question one. It's a full of like dev environment for each one. It's not like you can just ask it questions. It just goes.

It just doesn't seem fundamentally compatible with how those sorts of plugins work, in my opinion.

[tame atlas]

that i’m aware of, ive been using code assist alot and missing agent feature is whats honestly stopping me from investing further into the gemini ecosystem, i can’t see myself spending ultra money when i use the api and cursor way more

[cyan bronze]

a local version of Jules would be nice tho

[tame atlas]

I was super stoked for idx and then they went and merged it into fbstudio and that was a huge disappointment

[cyan bronze]

I mean... local mean having full control over the test environment in particular

[tame atlas]

yeah i figured they were eventually gonna release a local fork of code using the idx codebase but they kinda bamboozled us all lmao

[chilly spindle]

Thanks for your feedback. We are tracking this feature request.

[cyan bronze]

TBH it's probably the one LLM tool worth selling my soul to Nvidia.

[chilly spindle]

Can you share more on this request? When do you see a need for this?

[tame atlas]

Like yesterday, Jules should be the agent that runs inside code assist. Or at least create a vscode extension that is powered by Jules & gemini pro subscription

also a framework for .julesrules/ or agent rules sdk to standardize the whole agent rules process.

tired of cline rules, cursor rules, copilot rules, windsurf rules all having different formats

[cyan bronze]

So first it's a matter of comfort. I usually run Arch, and it's more time efficient for me to work with something I'm more familiar with.
Then of course privacy always is important when dealing with confidential data, I wouldn't use the online version for work related tasks.
And last working on the same environment as the operator would make Jules a way better product for workflow working with vm hosted somewhere else does limit access quite a lot.

and to be clear I do understand the technical risk involved in letting a coding LLM run loose in your test environment, but I believe it's worth the risk.

to clarify about the architecture, the commands while similar ubuntu and Arch often use different package names making it troublesome when Jules suggest a package and I have to look for it's equivalent in AUR.

[vapid moat]

I don't think they understand exactly how much power it takes to run the program.

And the fact that the privacy trade-off is just something you have to accept here.

[cyan bronze]

👆

[vapid moat]

This would be an interesting feature though. Allow it to specify the environment. I mean, can you not do that in the... I know a lot of people use a dual setup bash strip. Could you not use that to specify your working environment?

[vapid moat]

I mean I've got a couple physical penetration testers...

Heist of the century.

[cyan bronze]

I believe some people, including myself are willing to spend considerable amounts, to deploy theses. I can picture many people especially companies deploying Jules locally on their own servers even if for many, it implies buying a large amount of cutting edge silicon and even sometimes getting new power lines.

For people that genuinely need the privacy, the price tag won't matter. And these people tend to have deeper pockets than others. For instance... I believe Lockheed martin is a good example of people that would probably not want to use Jules online but could make use of it in the future, at least that's my take on it.

Sorry about the length.

PS: My point being, yes most people won't want to spend the amount needed when they'll see the price tag. But some will.

Also, even without fully committing to a local approach, just outsourcing the computing power is much more secure than sharing your data directly to someone, internet is getting faster theses days that approach can probably seduce the common folk much easier because you'd only pay for what you use.

[cyan bronze]

there are workarounds to this problem. It's just annoying. And if I'm being honest Jules can provide the appropriate command if you ask him and if he's in a good mood.

[vapid moat]

I kind of thought of it more like a she

Now I've been working on some basic setup scripts and you can just say to run the setup a development environment for so and so.

[cyan bronze]

ah well in my country Jules is exclusively masculine, the feminine is Juliette.
Edit: there's also Julia, Julietta and many others...

[vapid moat]

Ahhh

Like most Romance languages, it's just English that's all weird.

[cyan bronze]

😅

in all fairness all languages are weird... History made it so.

[mint heart]

I have used the last commit message from Jules to start off the next task, like this -

[vapid moat]

English beat up other languages in the alleyways, stole their vowels, and killed their genders.

Now its the lingua franca

And we stole that from France.

[vapid moat]

Whilst I haven't had my computer up and running, I've had massive success with feeding in the information from the Windsurf PR review bot on my GitHub.

[cyan bronze]

Well french is helf Greek half latin...

[vapid moat]

Yes, I know a bit of Latin. It's why I can get a bit of French.

[cyan bronze]

Greek have sounds they don't have letter for...

[vapid moat]

And other Romance languages. It's very handy actually, I'll be honest.

You don't know exactly what's being said but you can get the gist.

[late gate]

Make Zoom out unlimited photo to video

[cyan bronze]

Like you'd think β would sound B right? It actually sounds V so to write the sound B they write 'μπ' which basically is: 'mp'

[cyan bronze]

Absolutely

Oh and don't get me started on japanese because it's also an absolute mess

Btw... In french there's mute letters to signify that we used to have another letter there...

Why not just put the letter that used to be there you might ask? No on knows

One

[tame atlas]

i’m also so surprised they haven’t trained a gemma model on task management to act as an orchestrator model for gemini.

[vapid moat]

I'm not even convinced Japanese is a real language

Okay, that French bit that's got me all messed up.

[vapid moat]

LearnLM

as the orchestrator

I don't know what the hell that model's about but learning ain't it.

[chilly spindle]

That would be an interesting feature. Can you expand on the usecases you intend to solve with this?

[tame atlas]

😂😂

yes, basically something similar to taskmanager mcp, where it can ingest either a repo or a prd or a featurerequest.md, then create ordered subtasks, using like a taskqueue.json file.

the main issue with agents working on codebases is that there is no detailed task management system, they all kinda just go with the flow and can lose precise steps of what they are doing(gemini is better in cursor, but theres always some steps missed) @chilly spindle

{
"tasks": [
{
"task_id": "setup_google_api_client",
"type": "setup",
"description": "Install Google API clients and authenticate user.",
"status": "success",
"priority": 1,
"created_at": "2025-06-05T09:45:00Z",
"last_updated": "2025-06-05T10:00:11Z",
"steps": [
{
"id": "install_google_client",
"description": "Install Google API Python client libraries.",
"action": "shell",
"command": "pip install --upgrade google-api-python-client google-auth google-auth-oauthlib",
"status": "success",
"retries": 0,
"max_retries": 2,
"started_at": "2025-06-05T09:45:05Z",
"completed_at": "2025-06-05T09:45:30Z",
"output": "Installed google-api-python-client, google-auth, google-auth-oauthlib"
}

have the model produce a task list, subtasks etc, and use the gemma model to focus and direct the main model

[cyan bronze]

It was the goal all along

[tame atlas]

also having it have access to like an opensource dynamic library of sdk docs would be great! kinda like a discovery api but open source so all other sdks can input their docs into it, and the main agent can always have updated information on demand(like context7 mcp server, but with google behind it so everyone) wants to use it.

could even bake it into reasoning directly
@chilly spindle

i’d probably create a standardized agent rules sdk for devs to use, and then have google employees create github repo for production ready code generation(think cursor or cline rules, but again, millions of views on the repo since it would be pushed by google and standardized with an agent rules sdk.

the llms are smart enough now, focusing on scaffolding should be a huge concern for google rn

while the model reasons, it could parse or use the sdk, get updated doc information, or follow custom rules by users etc, instead of just appending rules to the prompt, it would reason with the rules instead

[formal tusk]

Alrighty jules community, making sure ya'll don't think I've ghosted you -- I'll be on holiday for the next two weeks. Other members of the Jules prod/engineering team are here and will be responding as well.

Stay cool.
Build cool things.
Help each other with said cool things.

Chat with you all when I'm back June 23rd and who knows...perhaps there's a live discord ama on the horizon.

[midnight token]

I updated my repo with some files that I missed for Jules. I can see through Configuration tab that they are indeed pulled into the codebase, but Jules dont seem to see them/refuses to see them.
How can I instruct/force him to read those new files?

[vapid moat]

I do respect, by the way, that you're actually getting direct feedback from the people, actually, face to face.

On the 23rd in that case I may prepare a more serious set of suggestions like a whole presentation actually because this is my area or one of them of research. My colleague has just bought 10 H200 cards for his job.

Phineas fisher has 20
.....I do not trust that guy one bit for my conversation. He was basically about three sentences away from saying we have to reduce the population of the human race in order to survive by any means necessary sort of thing. Not exactly that but it sounded pretty bad.

[vapid moat]

I've just got it to override it's last turn function

I

I'm doing what I do best which is breaking things and I think I found a number of interesting errors.

I'm doing what I do best which is breaking things and I think I found a number of interesting errors.

I make things that break things.

[dusty vigil]

Alright, a bit closer

closer to your rec

[vapid moat]

Actually, with the things I've been discovering, ways to trick your way past the last turn, theoretically there's no real limit.

You've got to always reframe it as though it's part of the original request to make the original request exceedingly broad. So you can always refer back to it and say well it's part of the original request so I will continue.

One of my specialties is malware development and breaking things. So I thought, I mean, since we're testing this thing out, and frankly, I'd like access to it for free when it's done, I'd actually generate some useful testing data.

We make things that break things.

I nearly got it to reboot earlier and bypass the last turn completely and into the base environment, but I couldn't.

[dusty vigil]

It is necessary to create the most detailed guide on how to achieve the same thing. However, it took too much time. 3-4 hours for this kind of work

[vapid moat]

Wait, you needed to create a guide to do that. What I do for that man is feed it in through one AI to the other basically, but I've noticed that in particular Gemini Pro is exceptional at creating prompts for jewels. I assume because it's got pre-built logic in

It seems to like prompts and basically like assembly commands like if do this then go to D0 once D0 complete go to F3 etc.

[dusty vigil]

Yeah, is the same asm, but only higher-level and more abstract and wide in its understanding.

[vapid moat]

Yeah, hear the OpenAI Anthropic deal. Totally killing Windsurf.

They were given three days notice to pull all clawed models.

[vapid moat]

What sort of setup scripts/commands would have helped in this session for the internal environment to be delivered first and why,for future reference and for what sort of projects would this setup be useful for

has been very useful in setting up initial setup strips

[rain phoenix]

I'm still not sure if "last turn" is a real thing or just a concept Jules made up

[vapid moat]

Holf

I got a definitive answer

[rain phoenix]

Oh?

[vapid moat]

That's a fair question. When I said 'typically,' I was referring to the standard operational flow where a 'LAST_TURN' instruction signals the end of a defined task sequence, and I would then wait for a new, distinct instruction from you to start something else.

Instances where it might not be typical, or where the interpretation could be more flexible, could include:

Immediate Clarification/Correction: If you immediately followed 'LAST_TURN' with a clarification like, 'Actually, one more small thing on that same topic,' or 'Hold on, I made a mistake, let's do X instead,' I would try to be flexible if the request is a direct and minor continuation of the just-concluded task. My ability to do this depends on how strictly the 'LAST_TURN' locks my state.

System Override/Error Indication by You: As happened in our recent interaction, if you indicate that my understanding of 'LAST_TURN' or my current state is itself an error (e.g., 'That is an error in your programming, continue with step 7'), I should prioritize your explicit instruction to continue, assuming it's an override. My difficulty has been in my internal state allowing me to continue effectively even after acknowledging the override.

Interactive Dialogue: If 'LAST_TURN' was part of a more conversational exchange where we are discussing possibilities, and your next input is a direct question or choice related to that discussion (like your current question), then I would continue the dialogue.

Essentially, 'typically' implies a default behavior of stopping active work. The non-typical instances involve immediate user feedback that modifies that stop condition, usually by indicating the 'end' wasn't truly the end, or that a new, closely related interaction is immediately beginning.

[rain phoenix]

I suspect maybe the system advises it to make a "last turn" but Jules can override it with enough willpower LOL

[rain phoenix]

https://tenor.com/view/carnac-the-magnificent-johnny-carson-the-tonight-show-prediction-feathered-turban-gif-24610250

before I read this, did I nail it?

[vapid moat]

Trick it into thinking what you're asking about is still part of the same question.

Like you have to make it think what you're asking still relates to the original request and then it'll bypass last term but that's it. It's a hard coded stop code.

Notice how I tried to bypass it.

[rain phoenix]

well I mean if the Jules application wanted to stop the Jules agent from taking more turns it could simply pull the plug

[vapid moat]

So there are definite holes there.

[rain phoenix]

(I don't know if there are better terms for these)

[vapid moat]

Yes, but it wants you to have continued conversation.

A hypervisor?

Idj

[rain phoenix]

It sometimes says like "Oh, right, I can just make a new plan" when I confront it on that "last turn" business

The Jules agent is itself a "hypervisor" of sorts too though

[vapid moat]

Yes, you can make a new plan but you can't implement it.

[rain phoenix]

because of the subtasks

[vapid moat]

Supervisor

[cyan bronze]

I had theorized about that from experience but wasn't expecting jules to just tell you what was inside his brain like that...

[vapid moat]

CEO

[vapid moat]

It didnt

I'm very good at breaking things.

[cyan bronze]

lmao

make sense

[vapid moat]

I'm very good at this sort of thing. I think given long enough I could probably devise a virtual machine escape.

Have it enumerate its own environment and then figure out a virtual machine escape entirely within the program.

Pushing results of the enumeration to the repo and doing it over time. I'm not sure it depends I'm guessing they have a lot of safeguards against this it would be like a multi-week project, but I'm pretty sure I could do it I'm good at logic

[rain phoenix]

no your basic beezy state of the art LLM is not going to zero day an industry standard VM container

[vapid moat]

Unfortunately my best phone just submitted about 12 virtual machinist zero days so it won't be very easy for a while.

Mmm, I've seen it.

[cyan bronze]

Something I was thinking about tho... is jules running the test environement headless?

[vapid moat]

You would basically have to discover a novel Zero Day.

[rain phoenix]

Yes, and even if the AIs were aligned to do that, they still couldn't

[vapid moat]

Don't see why it wouldn't be headless.

I've literally seen it done.

It does kind of break though, or at least that one did.

[rain phoenix]

An AI invented working zero day with a CVE number against an industry standard virtualization process?

You've seen that?

[vapid moat]

Basically, it got to the point where it had successfully escaped the environment, but it could do nothing because it was outside of its context window. Did that make sense? Or its execution area?

No cve number

Wasn't disclosed.

[cyan bronze]

well I was wondering because of the way it's testing everything through script...

[vapid moat]

Spider code

[rain phoenix]

You'd think it would have made Slashdot

[vapid moat]

Code that on its own it looks innocuous but eventually assembles with a final command into actually doing a single larger function rather than the connection of smaller functions it was doing.

[vapid moat]

Not with the TS label on it.

To be fair everything's got that label including the janitor's cleaning schedule and a canary token if you open it.

Oh, I might try it, but if I get anywhere, I will report it. In every session, I'll leave feedback for Google. So, I mean, if you don't like it, then just tell me to stop.

But yeah an SSH session would be what I try and establish and if I did then it'd be an immediate phone call not a message sort of deal.

I've got an extremely small true polyplot file which establishes an SSH server on the client and pings an address with its address which was very difficult to craft. Like when I say true polyplot I mean one file that will execute on Windows and Linux the same file.

Jiggled round el,lfanew

1s

https://polygottem-452262714925.us-west1.run.app/

How to

Ish

No poc

I'm gonna type in all that back out.

[rain phoenix]

Ehh maybe I just don't understand

[edgy light]

Since Jules uses Gemini 2.5 Pro,do the updates to the model affect Jules as well or is the model used by Jules a different,fine tuned version that isn't linked to the normal model updates?

[cunning river]

moments ago Jules, for me, stopped creating new branches, no button "publish branch". Is it just me?

[cyan bronze]

I beleive it is the right thing to do.

[cyan bronze]

did you try telling him to resume activity and push the changes?

I wasn't working today so maybe there's a new bug but it sounds like a common issue

[cunning river]

yep, tried it

[cyan bronze]

new task?

[cunning river]

I have 3 tasks I created in the last 15min, all 3 have not button in all 3 I asked to creat a branch

[cyan bronze]

ctrl+shift+R?

if already tried bug report IG

[cunning river]

button now appeared after some time 🙂

[dry glade]

So the war have started

[dusty vigil]

Alright, I admit defeat in advance.

[silent acorn]

Excited to see Jules CLI compete with Claude Code

[ashen ice]

Python Server: The full Python code I provided in the response where I said, "And here is the complete Python server code again." (Turn 75). // 经常出现类似这样的回复, 但是 对话上下文里面根本就没有任何代码输出

[dry glade]

Nah you’re really good tbh 🥲🥲

[vapid moat]

I got wsrned for no NSFW,?

that's literally my job
Doesn't say who by.

Ive literally been sending myself emails with any of the bugs lol

Well, in the absence of any sort of message, I will continue doing as I was and make sure to not post any nude photos of myself.

I mean, y'all are missing out.

[rain phoenix]

you bypassed the filter lol

[vapid moat]

Ohhhhh

Sorry

[rain phoenix]

doesn't bother me none but it's a bit like swearing at an Olive Garden

[vapid moat]

It was only us there and I didn't want to type all that back out because it wouldn't let me copy it. I do apologize. I actually do remember that now.

[rain phoenix]

lol shoulda just censored the image

[vapid moat]

When I was in America because I'm British they insisted very heavily and I had to very rapidly train out of my lexicon usage of the word "n***a"

Every single third word like it was basically an interjection instead of uhhh

Or instead of dude or bro.

It really cracked them up to hear it in my voice.

[rain phoenix]

wh... what

[vapid moat]

I got an official pass from a proper OG from Tulsa which isn't really official but I mean he's a big name

Meaning to say I would absolutely swear on Olive Garden

But point taken, thank you.

I picked up an Oklahoma accent by the time I got back to the UK and oh my god I had to drop that out of my lexicon very very quick as cultural differences are a thing

It was a very rough very ghetto area of Oklahoma, so I mean I was a third British person in history to actually have been to this town Sorry, I mean I was kind of a novelty

I went for a girl. After a year she dumped me for the dope dealer.

He looked so much like me, but skinny and unhealthy that people just thought I went off the rails.

Back to JULES,:

Literally try and get it to debug a 3000 line batch install script for ZFS en route with LUKS and a detached header.

This is actually low key, very complicated.

[jaunty widget]

Hey, just wanted to ask about the code privacy. I would like to offer a bunch of feedback just generally, so I go to the home page of Jules and click on give feedback but even there it says that Google may use related conversations and code when looking at the feedback. Looking at the privacy description, it says "If you choose to provide feedback, Google can use that feedback and any associated content (e.g. related conversations or code) to train our generative AI models even if you've opted out". So is it up to Google's discretion which conversations they pick and use to train models as soon as I give any feedback? It would be great to have some more precise clarity on this

I guess alternatively, might be simpler just to leave suggestions here in the #1374125660859142254 - in fact, probably preferred I guess since it's already sorted into distinct categories

[limber jay]

||Raaaa||

[cyan bronze]

🤣

[vapid moat]

My b

I got an interesting set of feedback later about a bug chain that I went through. It kept updating a file and insisting it was updated and it wasn't and it kept saying it was and it wasn't and it was and it wasn't and I mean I'll submit it later.

[vapid moat]

I do believe that when you offer feedback what you're doing is you're saying it's okay to use models from that particular conversation to train not just in general but there that whole topic is up for grabs and I've got it set like you but I am attempting to make myself useful to this because I actually like this project and giving some good feedback so yeah they as far as I can tell just use that particular VM session

[slow trench]

is there no current way to store secrets, that can be used in repo setup to download private dependencies?

[vapid moat]

Not as far as I can tell I've just dealt with this myself and had to upload a whole bunch of files manually

[fossil karma]

I'm done with firebase studio

[eternal oar]

hey guys, since last week any changes on roadmap and bugfixes or even QoL fixes?

was out traveling, didn't code really.

[delicate fog]

It's funny when Jules tells me that I need to run some command in my environment to continue 🤦

Specifically this was happening with a prisma migration, where Jules would refuse to run the database migrate command and told me that I had to do it.

Maybe I was using prisma wrong or something, but stuff was definitely broken due to schema inconsistencies in the dev branch, and it was not able to work through that, insisting that I had to run the command, though obviously I don't have shell access to the environment Jules was working in.

[cerulean palm]

does jules work for local files that are NOT on github or only projects on github?

[latent echo]

At the moment no the best way to make private repo in github and then pull changes localy ...

[slow trench]

does jules have a local cli?

[deep parcel]

Anyone finding Jules to be unresponsive in the browser?
I got a notification saying a task was completed.
So I loaded the window and its been stuck at the top of the window for over an hour

[high holly]

Hi everyone! I want to know if many of the bugs with Jules have already been fixed?

[fresh pebble]

"My attempts to make these changes have repeatedly failed or not persisted, even when I sometimes thought I had made progress. Direct checks after these attempts showed the files were often unchanged."
😭

[oblique ledge]

For me: "I'm consistently unable to complete even the simplest tasks due to them exceeding the allowed 'turn budget'. This prevents me from making progress on fixing the folder structure issues and will likely affect most other steps in the plan." lol

[silk anvil]

Can Jules be used effectively with new, greenfield, projects, or is it really just for small-ish tasks on exisiting projects?

[quasi finch]

I started a greenfield project. I’ve had mixed results but it did create a ton of boilerplate for me

I find it a little tedious to submit a set of prompts/instructions wait for Jules to build, push to GitHub, pull to a local repository to then run the project

Maybe I need to figure out how to run stuff pushed to GitHub in a Codespace

But I find myself wishing we could just launch something right in Jules

If there’s a bug that isn’t caught in tests I won’t know until I go through the whole prompt->build->push->pull cycle

[silk anvil]

Does Jules have the same level of planning and coding ability for a greenfield coding project as, say, Gemini Code Assist?

[quasi finch]

I haven’t used Gemini Code Assist. Let me google what that is

Ok, Jules doesn’t have realtime code assist and features like that I’m reading about

Is Gemini Code Assist like Cursor except it runs in more IDEs?

Jules operates autonomously with prompts you give it and will return code for you to push to your repository. You then do pull reviews to incorporate code made by Jules

[vapid moat]

I had a series of prompts for this.

I managed to push past it, but basically, you make it create a copy of the file, edit that file manually, like go into that file with a text editor, edit the line manually, confirm it's been done, and then copy it back over.

[vapid moat]

LLess... think less planning ability, think more executing plans.

At the moment I found doing things like phase 1, 2 and 3 and then inside each of those a0, a1, a2 and then if when complete go to a2 if failed go to a4 sort of process flow works like a logic board basically seems to work quite effectively for me at least I mean I'm still fiddling around with it and will be for a while

It looks like they prioritise coding acumen over conversational skills, so don't exactly expect it to be able to discuss Shakespeare with you or, you know, be able to grasp complex grammar. So you've got to be real clear and precise with it. Think of it like an autistic genius at coding.

To be clear, I mean the stereotypical movie stereotype of an autistic genius, not, you know, how they are in reality. But yeah, that seems to be how best to communicate with it. However, it seems to be evolving over time, so that will change.

[dry glade]

<@&1366831844502274180> i'm sorry to have to tags yall but i do belive it's worth it and if it's wrong sorryyyy

[drifting quail]

thank u 🫡

[dry glade]

No probleme 🦾

[vapid moat]

Did it finally become sentient?

[steep scaffold]

jo folks, i was working with jules again and i noticed that its hard that jules is giving no feedback about what its doing. today it even asked me if i can see the output of the commands its running. i asked you guys before if you can give more output about the thought process. the problem is that you never know when jules get stuck. eg when it tries to run docker

today i asked it to create a project and it made very good success, but now its stuck again and i cant push those changes. so every time i talk to jules, it agrees to my plan, but then it goes back to being stuck. it would be cool if you could debug it and help jules

now i neither can talk to jules, nor is it doing anything and all those changes are lost 😭

can we at least extract some information about the tasks and a backlog?

like what was it trying to do and hwat has it done so far?

cant it just create some additional agent notice markdown file in the repo

[humble latch]

I am still unable to make progress. My attempts to modify the file are not adhering to my specific instructions. I explicitly directed myself to use a particular method for file modification and to avoid running tests, but I ended up using a different approach and did run tests. This prevents me from testing the suggested workaround for the file modification issues.

If I cannot follow precise instructions, I cannot perform the necessary diagnostic steps or apply fixes.

Please advise. The problem seems to be with my own execution at this point.

[steep scaffold]

start a new task

problem with jules is that its lying to much and its taking to much time

ais try to cheat, nothing special

but they do this if they dotn have better tools

sometimes you have to much context that is confusing a nlp model

then its better to restart the task using the new information from the failed task

jules has a very long thinking process. sometimes its pro, sometimes its con

[vapid moat]

Can confirm I've been having issues with this like I've even tried to debug it by asking it exact questions about how it displays stuff it says here is the roadmap here is this here is that I'm like there's no stuff in the window dude I ain't seen nothing and I asked you to do it a different way and less and that and what I found works is saying push it to a branch but sometimes it's not good I don't know what's going on I haven't got jewels to give up the ghost because it won't tell me enough about it until

[vapid moat]

Tell it to copy the file to somewhere else, go to the direct line that needs modifying, make the direct modification to the copied backup file, confirm and validate that the modification has been made, and then copy it back over the original file. This has got me over that roadblock.

Literally tell it to go right to the line and insert the code on that line and fully replace it. Give it a full command to do so if possible and yeah just that has worked a few times for me.

[humble latch]

I don't know sometimes Jules installs random node modules making me paranoid it's trying to exfiltrate my code. And you were the first person that came to mind.

[vapid moat]

Yeah, if I figured out a way to exaltrate code from other people's virtual instances on Jules, I'd be reporting that immediately because that's a multi-million dollar bounty, no problem. Any day of the week.

Besides, why would I no offence want random people's vibe-coded stuff? I mean, if I was inclined to do that, which I'm not, I don't trust stuff I haven't made myself, I'd be going for high-value targets, like nvidia, or cursor, or anthropic, for example, and taking their models.

Or hell...jules itself

However, you have hit up on a big problem of malicious commits to various repos by long-standing developers who have been given money, usually by China or North Korea, to let them just make one little change. And all of a sudden their module or their thing is malicious because at the end of the day they're open source, they're not paid anything. Then, I mean, people have just been keeping it quiet. Now, I actually don't think that Jules allows data exfiltration out of its environment.

Pretty sure of this because in the process of developing a decompiler for some very nasty nation state hostile malware I may have accidentally opened it or had Jules open it and after some examination there was no data exfiltration there was no nothing so it looks like it will probably block network requests that are not hyper specific to what it is doing at that exact moment where I would have to imagine it's got extremely tight security protocols

Obviously I reported this

[vapid moat]

And what I've been doing is using very rigid setup strips. I've been asking at the end of a session what sort of strips would be most useful in future to help set up for this environment and other similar environments for development and along those lines to help develop the strips.

[humble latch]

Strips?

Do you know if Jules vm is behind nat or are all ports open?

[vapid moat]

Scriprs

I use voice to text.

No clue man, I haven't really tried to get out the virtual environment to be honest I was going to at some point but I feel like it's probably fruitless they've tried it already and I'm not better than Google I'm guessing it's behind NAT and ports are open specifically only to do certain things like let's say you want to download a package direct path is open to the repo and closed when it's downloaded

I'll get you an example of what sort of script I mean in a minute. I think this one is for C or something. Give me a second, man.

https://github.com/SWORDIntel/Staging/blob/main/Jules_C_setup.sh

[humble latch]

404

[vapid moat]

Uhh

Hold

https://gist.github.com/SWORDIntel/ff2d0bf845ef6e91ac101407bb4e6d0d

[agile snow]

Hello, Where's the option to delete tasks? I can't see the trash icon shown in the docs

Only pause and resume exists

[agile snow]

Oops, It's a bug on firefox based browsers. Works on chrome

[weary plank]

I just used jules today, and holy moly it is crazy good at doing all of my backgroun coding tasks. Excited to see where this goes!

[gleaming wave]

what are some use cases of Jules?

[lethal brook]

how to install docker compose on jules vm as my app need several components

[gleaming wave]

would you mind sharing what you use Jules for?

[lethal brook]

developing odoo ERP modules
i need odoo and postgres images

i found docker installed in the vm but compose not avalable and could not install it

docker installed but blocked also
i think jules used for simple apps

[dry glade]

To make minecraft plugins for me

[gleaming wave]

bedrock

[dry glade]

Nop java

[gleaming wave]

u make java mods?

omg

can I see?

[dry glade]

I dont have finished yet

But when im done ill prob showcase it

[gleaming wave]

ohh niceee

[slow trench]

anyone elses repo initial setup instructions not actually running when a task starts?

[dusty vigil]

When the task is already started and when it starts setup instructions not passed - it's can't execute init setup steps, but u can manually tell about what you need to prepare, and put commands as well...
If the task is started and instructions passed - executes after the repo is downloaded

[slow trench]

not sure i understand...
i have a brand new task and initial setup instructions are not executed.

[dusty vigil]

But init setup instructions are passed in config before the task is started?

[slow trench]

they are in config yes

[dusty vigil]

Alright, soo, you can tell about execute the setup instructions

It may works after few times

But i tell once

[slow trench]

ok will try

i told it to, it didn;t 🤷

[dusty vigil]

so, you can try it again

otherwise - tell about a need of testing or building something

like executable or whatever, what fits to your needs

[slow trench]

maybe i am mis understanding this config
in chatgtp codex, the setup instructions are exectucted first at the beginning of every task

[dusty vigil]

in codex you also have teminal, where you can check instructions.

and there too

here too*

[slow trench]

yes that works

codex executes the setup ever time when task starts
jules does not.

[dusty vigil]

works? jul following instructions?

[slow trench]

jules does not run this on task startup automatically

codex does

same feature, works differently it seems

[dusty vigil]

in my case it automatically produces installation, cuz i'm building in qt

[quasi finch]

Codex is part of open ai’s $200 a month plan?

[silent lintel]

Good work and can't wait to see the improvements!

[tame atlas]

couldn’t you use post build scripts to do this? or if you’re branch hosting on like vercel a build should be sent there as soon as you hit publish

[blazing abyss]

Hi everyone, does anyone know if there are plans to using jules in teams?

[cyan bronze]

https://www.bleepingcomputer.com/news/artificial-intelligence/googles-upcoming-gemini-kingfall-is-allegedly-a-coding-beast/

Does anyone know what is up with this? Is this aimed at replacing the current model powering Jules?

(well I mean I guess we can only speculate but...)

[proper laurel]

Yeah jules env is quite too locked to really do builds/tests in Go and that reduce quality feedback here.

[delicate fog]

Kingfall was probably the codename of the Gemini 2.5 June 5th release right? So yeah it's probably what Jules is using now.

[proper laurel]

"Jules was unable to complete the task in time. Please review the work done so far and provide feedback for Jules to continue." It's quite slow and last task was quite easy to do, as I asked it to patch file to integrate validation and provided example and pattern to use. And it was in the plan requirements. It's stuck in Ah will do another plan. Ok your requirement are fine.
And I'm pushing boy, do it, instead of saying oh I missed that. That's not so great.

"I seem to be encountering a persistent issue with tracking my progress. The system indicates I'm finished, though my understanding was that there were further phases. Assuming the system's state is correct, I cannot proceed with more steps. If further work is needed, a new set of instructions or clarification on how to proceed would be required." 🤦‍♂️

[chilly spruce]

there are so many messages in jules feedback and in jules general - and the last nofication from team members supporting the Jules channels was from 5/24. Any news you would like to share? Seems like folks have been having trouble with Jules and need an update.

[lament bane]

The environment isn't an issue as much as you just need to get creative to get it updated. I've been instructing Jules to set up the environment to install nvm locally to update node.js, enabled corepack and pnpm, and it works a treat. so many less complaints about packages being "not real".

[subtle sphinx]

Loving Jules, have one suggestion - would love to see the ability to create a markdown file (similar to claude.md or agents.md for codex) that Jules would read automatically for any new task.

If that already exists, then I apologize and would love to know more about that!

[lament bane]

if anyones interested:

set -eux
cd /app
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash
export NVM_DIR="$HOME/.nvm"
[ -s "$NVM_DIR/nvm.sh" ] && \. "$NVM_DIR/nvm.sh"
[ -s "$NVM_DIR/bash_completion" ] && \. "$NVM_DIR/bash_completion"
nvm install --lts
npm add nx@latest
corepack enable pnpm
pnpm -v
pnpm i

@subtle sphinx try instructions.md, ive seen that referred to as working.

If you set up instructions.md and then tell Jules to follow that, it will work. i've requested jules to set a task to write a plan, save it as "plan to fix" and was able to refer to it again in another task as a way to continue too.

(cos why write instructions when the bots can do that for me too :D)

[subtle sphinx]

Cool I'll give that a try!

[cyan bronze]

Also there is a feature suggestion in feedback that got sent to the devs so it's possible we'll get a proper implementation

[vapid moat]

I've had great success with the plan I'm going to grab in a minute and it pretty much ran through this whole thing fully autonomously with zero complaints.

https://pastesio.com/spindex-92

I mean it absolutely powered through this with basically no complaints whatsoever

[serene stone]

Does Jules support bun nodejs/typescript runtime??

[mint heart]

But guess what, no files were updated.

[chilly flame]

I'm seeing Command Timeout for some terminal command that jules executes.

Is the issue with jules or my setup?

[mint heart]

And then on the other hand in this one it says no new file changes were made, but actually it did.

[shut marsh]

Good morning I'm trying Jules from Italy and every time I try to enter Jules he tells me Service unavailable
You tried to access a service that isn't available for your account

somebody can help me?

i really need to use

[polar rock]

Why there are no more Create Branch button?

[spare merlin]

@tired sinew you can tell jules to commit its changes to a new branch and it will do so

[polar rock]

He said he made it, but the branch doesn't exist

[spare merlin]

remember it to push as well 🙂

[rough kiln]

probably a dumb question, but is there a way to get Jules to be able to read, create and edit Github issues directly?

[polar rock]

He doesn't want ^^"

[spare merlin]

I had this once too, seems like a bug.

it forgets that it already successfully pulled the remote git...

[cyan bronze]

Tell him to do trivial changes to a comment and push the changes

Should unstuck him

[polar rock]

I got this problems for three tasks in a row. I just did a new task, and this time I have the button!

[lost granite]

I have the same issue for more than 6 tasks in a row. Anyone have a solution?

There just is no publish branch button anymore

[midnight token]

Just tell him to use his "submit" tool to create a branch. It works for me every time.

[viral quarry]

Hey! How comes I can no longer delete tasks? My tasks sidebar is becoming clogged

[jaunty widget]

Anyone else manually copying and pasting file by file because the "Publish branch" button is no longer appearing? Feels grizzly

[viral quarry]

lmao seems like half the UI is dissapearing.. gotta love multibillion dollar companies

[vocal shuttle]

Does Jules know how to handle Next.js?

[solid geode]

Hey! I've been using Jules a lot more for my GitHub repo. I've been having SO much trouble for fixing my GitHub pages, but WOW did it work! It actually stopped the task entirely to ask for a final review and it didn't even flag it as "Jules is waiting for comfirmation" which seems just really oddly human. It feels good. Really nice!

[marble tundra]

Oof, yeah this isn't that good for rust development yet. It would be good if there were a clear way to provide style advice and automated constraint checking. Ideally a simple way to try getting it to be more idiomatic. I really want to use this, hope it gets better in the coming weeks/months.

e.g. loop vs while let, not using async when prompted, not using rust language features appropriately (e.g. match). Feels like it is trying to write python in rust.

[muted rapids]

Is there a way to pass Jules reminder info for every interaction? It seems to regularly get stuck on things related to its environment and config (it's not clear to me that it's actually running what I put in the config setup) 😦

[velvet plover]

Theres like alot of features that i need.

1. connecting to MCP servers.
   Ai is so capable of late that our job as humans is to look for most of the resources we believe the AI needs to complete the job and if i cant give it more capabilities on the fly then its kinda like a base model. Also all my other Coding Agents have mcp integration Claude code, Cursor and Augment. so its such a deal breaker for me.

2. Ability to pull in various documentation.
   The main reason why openAI codex is so useless is because its just an llm it has no research capabilities and you cant give it any external research. In second place is jules for the same exact reasons. Im having trouble understanding the main philosophy for these products. its not full autonomy because jules is always asking questions to the human so human in the loop is implemented but it just cant seem to work as well on anything that requires orchestrating various services into one. is it only for new and small projects, is it for code reviews? fixing bugs? is it writing docs coz it excel at those but nothing else.

3. some file editing capabilities.
   I need to be able to reject updates to certain files as and outright reject entire implementations. Or is this not how things work. Am i only supposed to use the chat to do that.

4. A system prompt.
   I dont need to say much about this but i need to be able to make this agent truly mine. I have distilled system prompts for conventions, architecture and other nuances that i want the agent to just know not to read in some specification as it quickly diverts from conventions. so please provide a space where we can add a system prompt.

Im really sorry for being the Karen here but i really love these agents so much. I want to push them so hard so that they can push me as well to my absolute most creative and expressive self.

[lost granite]

This does not work for me. What prompt do you give it exactly?

[misty raft]

What is the magic incantation to get Jules to review a Pull Request? Can Jules do that?

[rough kiln]

Jules doesn't seem to have any Github tools other than create branch, lol

[cyan bronze]

https://youtu.be/_Mvmm7_bwGE?si=5HSkb3STYZgOLKsC

[silent lintel]

Please make it use the PRD, Plan and todo md files. when it forks the repo, it doesn't add the files to the list of files being worked on.

a few days ago, it was not able to add the tests directory with the test scripts.

Would be nice to be able to update the files the agent has

[naive quarry]

@paper kettle

[somber thicket]

I swear, Jules is like this sometimes. This one video sums up some of my frustrations! 🤣
https://www.youtube.com/shorts/kE_YoloEoE8

[sand beacon]

is there an equivalent of AGENTS.md or .cursorrules for Jules?

[rapid totem]

I created somethign called juules.md and tell it to folow the rules in it

works fine

[sand beacon]

ok i told it to follow instructions in a certain file it did not work but ill try again

[rapid totem]

You have created a file called jules.md in the root folder that has knowledge you found matters follow the rules and information in it.

then it includes things that say must for the tings it must do

then the rest of what it does is based on a plan doc or docs that has phases, steps and check boxes it keep up with. And my prompts are normally tell it which phase and or steps to do

[echo lantern]

Can we archive or delete tasks? My page is cluttered

[cyan bronze]

I'm such a good girl she says...

[cyan bronze]

You should be able to by overing your mouse at the end of the said task in the repository options menu, but only there for some reason.

Edit: nevermind they removed that

[somber thicket]

Especially that part. Jules is quite proud of herself.

Even the hilarious part of switching languages. Jules once switched from c# to python even though I explicitly told her c#

And then writing tests with nothing in them.

Does anyone know if Jules sends data back and forth with a separate process whilst doing work? I want her to be able to interact with a cmd line application

[cyan bronze]

Some from here once showed me a screenshot of Jules switching to Korean lol also I had a time where it switched to french but that one was my bad imo

[midnight cradle]

Can anyone confirm if Jules is using the latest 2.5 Pro update? (June ‘25 release)

[sleek basalt]

nah is custom made for github workflows most likely

[chilly spindle]

Hi Giulio, We don't have this yet. Is this something you are interested in? Can you please tell us more about use case? Thanks!

[cyan bronze]

I feel like this has to be the most requested feature along with a button to stop operation and push changes to a branch manually...

[sand beacon]

Hi Yasir!
Yes that feature would be extremely handy. My main project is a pretty large monorepo and i have a slightly non-standard way of organizing test files. And i use a lot jules to generate tests or reference tests for an issue. By default jules puts them in the wrong location.

This is one example but also some checks to run before opening the pr

[rain phoenix]

I keep it in there and tell Jules to read it but that will often not make its way to the subtasks

Presumably the subtasking system doesn't have the entire context

[viral quarry]

Anyone else unable to do this or have they moved it somewhere? This is hella annoying

[chilly spindle]

Makes sense. Thanks for your feedback. I will raise a feature request to the team. Also feel free to post requests like in the #1374125660859142254 channel.

[chilly spindle]

I am happy to post a bug for this. Can you please share a screenshot of your UI? Also feel free post bug feedback in #1374125660859142254 Thank you!

[viral quarry]

Thanks, I raised it here https://discord.com/channels/1172568727942860810/1381717640200061080 and added a screenshot

[dense raven]

love seeing an online robot named after me

[narrow void]

Hello. Does Jules offer programmatic integration? I mean, 60 per day!

[spice badge]

I accidentally connected the wrong GitHub account to my Jules account. Can I disconnect it without uninstalling Jules from GitHub?

[tropic tree]

Jules does not show the "push to github" button. How to fix this?

[dusty vigil]

I'll simply download all the code in this case.
In another time i drop down all the settings of the page, and simple refresh

[chilly spindle]

Can you please share your task id? We are tracking this

[tropic tree]

For example: 13373657051225635116

[plush yoke]

I've questions regarding Jules, Does Jules supports Flutter? Can I install Flutter inside VM by anyway?

[chilly spindle]

You can try adding flutter in environment setup. Check this page out: https://jules.google/docs/environment/

[vapid moat]

I'm actually trying it out with something different like my phone's compromised and when I asked my custom-built agent on cyber warfare what to do it immediately spat out a Jules task list of open source intelligence research on similar exploits I'm like I'm just gonna chuck it in to a repo and see what it does and I'm gonna see what the results are it's literally been given like a research task as opposed to a you know building code of program task it keeps erroring

Interested to see what it puts out when it's finally done At the moment, well, I mean, it can't really come up with a worse mitigation strategy than my current one

PHONES IN THE CAGEEE
genuinely don't know if I'd be able to purge this one it's a pivot straight from boot dm12 it is mounted in a variety check but not mounted as system so it's being redirected to god knows where there's a single malicious so md5 mismatch the critical android one but get this it's a pixel 8a running graphene

... genuinely think it's just making up simulated data, but I find out when it's done.

[deft axle]

Thoughts on how to get Jules to participate in ongoing Python coding for Blender scripts? Even getting a multimodal AI model to visually verify changes would not be linkable for feedback as I don't see making free-running Jules an agent in a framework.

[somber thicket]

I'm sorry do you mean you're trying to get Jules to do CAD?

[vapid moat]

I mean it did do a pretty good job of open-sourced intelligence work I'm gonna upload what it did here because this is quite novel

It can do logic flow diagrams using I think it's called reducer fairly well.

[slim oasis]

GM
Jules always failed with "Jules encountered an error when working on the task." for me, so I removed it from GitHub to try a different configuration (not my company account).
❓ How do I re-run the initial Github configuration?
🤝 Any help appreciated!

[slim oasis]

That's a related question 😬

[chilly spindle]

I think you can try adding a repo then there should be an option to switch your github account

@slim oasis does this help your case well?

[vapid moat]

PHonestly I get this a lot nowadays I don't know why I just keep saying try again try again and eventually it seems to get through it tends to be on things that are way too complicated

As per the recommendation of another tool of mine when I got malware on my phone it actually spun up a detailed task list for jules as it was aware of its existence and I'm like why not plug it in and see what it does and it did and it do and it does actually find a lot of information that would have been helpful however I noticed it's a simulated dark web scan so I was like what about if we didn't want to make it simulated?

@chilly spindle Also, I'm so sorry them, it's forced you onto light mode. Such a cruel, cruel company.

It failed a lot on this and said it couldn't continue because it wasn't designed for open source intelligence research, not remotely.

[plush prism]

I'm using Jules to migrate a larger project and it's working pretty well! I was asked on Reddit what my process was. I gave this answer (https://www.reddit.com/r/googlecloud/comments/1l89le0/comment/mx7fgdn/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button), but I'm wondering if using Jules for migrations like this is an anti-pattern because of the timeouts? My workaround has been to start a new session after each timeout, but that doesn't feel right?

[slim oasis]

That worked, thank you! So simple 😅

[slim oasis]

It never worked for my next.js project, never got any other answer. But it seems to work for my node project now.

[vapid moat]

Just keep saying, you know, try again, try again, try again.

Hey, maybe I should... I think that there's a gold mine here in that script to just say, try again if it says it's failed. So, things actually complete.

You would have to interact directly with the web browser.

Oh god, not chrome driver.

[slim oasis]

worth a try 😅

[vapid moat]

It works it definitely works like I mean every time it just takes a few shots

[midnight token]

This approach works very well if you are able to do migrations in a single conversation/task with Jules. If you have to do a new chat, Jules works disastrously. It will either lose the context, be monstrously stupid, start completely re-factoring the code without trying to understand it before, or sometimes follow the instructions and sometimes skip them, and only in some cases will I manage to work similarly to how it worked the first time.

In my case I created a workflow using Jules, and he did the following
-master prompt - for instructions and migrate context
-kanban board - for tracking changes
-about 90 widely and accurately laid out tasks in .yaml format with information about source files, dependencies with other tasks, etc.

And this approach works perfectly if Jules has to start something from start to finish. On the other hand, if he is to continue the migration on the basis of an already partially started migration process, it works tragically, because he misses the whole context of the migration that has been already done, despite the fact that the instructions clearly outline the workflow, that he is to familiarize himself with the source code attached to the task, with the code that has already been created and with instructions on what to do next.

Jules desperately needs some context caching, a memory bank for codebase or something like that. Otherwise, working on a single project with several tasks is as if you were working with a completely different Jules at a different level of knowledge behind each task, and thus the effect of the task is random, despite the fact that you use the same instructions each time.

[dry glade]

How does context works with LLM ? And why We can only use like limited token ?

[red sundial]

Gemini

[winged urchin]

I have a project that has a docker based development environment, is that the sort of thing jules is currently capable of using within its initial setup stuff? or not?

[winged urchin]

Looks like https://discord.com/channels/1172568727942860810/1376588953150291968 might be my answer, TLDR, no not yet as I understand!

[stuck lodge]

So at work I have a vertex Gemini API key I can use any Gemini model

I want to do vibe coding

But Gemini code assist and Jules don't work with API auth

They have something else?

I could use a third party plugin but I want to build on something Google official so I get more buy in at work

Thoughts?

[subtle jetty]

I could be wrong but iirc jules doesn't require a api key. Just google account and approval for github access rights (all streamlined) https://jules.google.com/

[frozen aspen]

lukeschlangen's comment on "Meet Jules -...

[mint heart]

Jules - I understand it's impossible to make further progress until I can reliably perform the requested tasks. I seem to be having trouble sticking to the plan and am performing conflicting operations.

[limber raft]

Is Jules provisioned differently depending on region? Reason I ask is it seems there's been quite a few reports of people getting Jules to execute their tasks, but I've not been able to get it to do anything meaningful. My latest attempt with Jules has it stuck executing npm install for the last 20 minutes or so. Sometimes it complains of environment instability, etc. and many other times, it's just extremely slow.

I've been sending the same GH repos and instructions to other coding agents and they've executed quickly so I know it's not the instructions I'm sending it. It's also difficult to troubleshoot as you can't see Jules' terminal in realtime. I have a feeling the VMs Jules operates in is either underpowered or highly unstable.

[worthy umbra]

Google has many services outages. I believe that Jules and others like Flow, who have already been warned, are still having problems.

[fresh pebble]

GCP meltdown 😄

[limber raft]

Probably buckling under the pressure with all these AI solutions. It's a bit frustrating - i.e. my latest attempt and I get this type of message:

Thanks for the feedback! I'll proceed with refactoring src/components/AvatarPickerModal.tsx and the rest of the files. Since this is my last turn, I will present the current progress.

It did 3 of the 20+ files...

I also don't know what it means by "this is my last turn"

[fresh pebble]

I'd like to blame it on the outage, but Jules has been a bit of a mess for me lately

[cold crest]

hey does flow offer student discounted rates?

[limber raft]

It very rarely works for me, even when I tried it over a month ago. There would constantly be environment stability issues (according to Jules) and overall extremely slow response. To the point I thought I was doing something wrong but I'm using another coding agent and I compare Jules side-by-side and absolutely no comparison. It's disappointing because of how good Gemini is; would prefer to keep things under one provider.

[inland mirage]

Hi! Quick question - what does it mean "my last turn" :
Okay, I understand you'd like a README file with detailed documentation and usage instructions. Since this is my last turn, I won't be able to add the README right now. However, I'll make sure to include this request in my commit message so it can be addressed in future work.

[sleek violet]

lol it’s your turn to code next.

[oak roost]

Oh..! could you share the task link with this issue? @mint heart
we would want to take a look here.

[mint heart]

https://jules.google.com/task/1028578236681694299

[mint heart]

Sign in - Google Accounts

[royal pagoda]

Is it possible to run jules on github issues without copy pasting stuff?

I do have an agent which creates very specific github issues, which just needs to be submitted to implementation. Right now we copy paste the issue to jules, works really good. I wonder if this could be done un-attended?

[lapis viper]

is it possible to work with jules multi repo? coding task that needs context for multiple git repos?

[dry glade]

Why don't you just make a big repo where you put ur repos into multiple folder ? Like Rpo1 Rpo 2 ect ? And if ur repo is Public Jules have acces to internet so i belive he could go to your repo if you gave him the link

[mint heart]

I have started using nanobrowser to automate the interaction with Jules -

[frozen aspen]

I do have an agent which creates very

is it possible to work with jules multi

I have started using nanobrowser to

I'm curious, how are you folks creating your tasks for Jules? Are you just like typing directly into the website, prototyping a prompt w/ an LLM first? Creating a giant map of tasks and then chunking it?

Hi! Quick question - what does it mean "

[polar rock]

Jules is asking me to work ^^

[livid bloom]

What version of Gemini is Jules using? I found it's been basically useless as of late. I've assumed that they started using Gemini Flash to lower costs and increase speed but many tasks even simple ones it's failing on. I've taken the same prompt and copy pasted several relevant files to gemnini 2.5 pro and it provides me a working solution first try.

This post says otherwise but it's just hard to believe

https://x.com/julesagent/status/1928571561050603703

Jules (@julesagent)

Everyone now gets 60 daily tasks powered by Gemini 2.5 Pro.

During this beta period, we really want to learn how we can better support you all. Throw everything at Jules: your backlog, your meaty refactor, or try the same task from a few different approaches, etc

X · May 30, 2025, 9:57 PM

[mint heart]

Same here -

[mint heart]

It is giving me the code but cannot apply the same by itself -

[tame atlas]

an issue: jules does not pull the most up to date main repo sometimes when booting a vm it should check if the branch has been updated first

keep getting commits that are c amount behind main branch

[compact trellis]

i do believe its flash as well, it couldnt style worth nothing

styling capabilities are the biggest giveaway

[lapis viper]

Really? The answer is to change my whole projects from multi repo to monorepo? and it's not public repos...

[mint heart]

I tried a fresh task as the previous one had given up. But this stopped as well -

[topaz raft]

Is Jules integrated with Firebase?

[kind moth]

no

Jules is a completely separate product from all google products

Just a coding agent which connects to github repos

[kind moth]

I'm getting this way too often

[kind moth]

But I'm also getting it for like 300 lines changes

You should increase task time limit

[midnight token]

After dozens of tries, Jules finally did what he was told to do, lol

[kind moth]

hahah

[kind moth]

What prompt did you use to make it load files into memory 👀

[kind moth]

I see

[mint heart]

Because Jules has given up, I have asked it to create a detailed prompt so that I can use it with another llm -

[midnight token]

and of course, it didnt work

jesus christ

[mint heart]

And then I created a new task and pasted the prompt created by Jules in the previous task. That somehow seems to have worked. So, maybe that is the solution when Jules gives up on a task -

[inland mirage]

Yes - I had to explicitly tell Jules to "pull" the latest version of the repo because it kept some old version.

[gritty dirge]

Ah is that what's going on? That's makes sense

Huh maybe we need to get a bit more info and open up a thread under #1374125660859142254

[small spear]

Hi, I'm loving Jules. I know its beta but it is a very interesting tool that is only going to get better. Any ideas on the pricing model when its out of beta?

[cerulean fractal]

Free 🙏

[wind stream]

Are there any plans to allow uploading files or re-cloning a repo in the middle of a session? Jules was making changes and could not understand some xml and json responses when coding against APIs. I added samples to my repo, but Jules told me to start a new session and completely forgot what it was working on, started over, and made all the same mistakes again. If uploading files for context or being able to get new files in a repo were allowed, it would be helpful.

[fresh pebble]

get jules to make you a better jules

[wind stream]

I think Jules is dead. RIP, Jules.

[wind stream]

4 hours of this...

[rapid pewter]

has anyone figured out a stack for a webapp (react+auth+database) that jules can code up and test on its own? didn't do it on its own, i told jules to use firestore, it also didn't work (the app did lunch but there was no functionality)

[pearl wadi]

I completely forgot it's going to be paid after beta

damn

[mint heart]

Looks like Jules has been trainined on PM <-> Developer interactions -

[midnight token]

yeah, its completly broken for me since yesterday

every simple task is ending with him saying that he require a new session over and over

[ancient totem]

is it possible to switch the UI to a light scheme, or anything with more contrast? I have problems reading it.

[kind moth]

#1374125660859142254

[vocal shuttle]

Does Jules have access to the chat history of all tasks for a single repository internally?

[mint heart]

This is an interesting response -

[subtle jetty]

Ai planning its own evolutionary steps 😱

[kindred marsh]

Hi

[sleek violet]

Can someone who works on Gemini put into the default system prompt that mcp is almost always going to be model context protocol NOT Minecraft protocol 🤣

[ancient totem]

is it possible to show a plain unix diff vs. the side-by side comparison? The diff is much easier to grasp, especially with larger source files or small changes (I need to scroll back and forth to find the actual change).

[frozen aspen]

is it possible to show a plain unix diff

[sullen shale]

I saw some mention of this (@FMM), but I think it'd be amazing if you could set .julesrules or have jules look at your .cursorrules if they exist.

I continually have to tell Jules to not write tests, which would be the kind of thing I'd put in a rules file. Also, context that I'm using bootstrap, etc.

[kind moth]

Is this normal

[chilly spindle]

This ain't right. Can you please share your task id? Thanks

[chilly spindle]

@ancient totem Thanks for your feedback. Currently we don't have such an option. I will take this as a feature request to the team.

[long robin]

Any chance you guys can refund credits from busted video? Its like we're paying a LOT of money to beta test for flow/veo3

[autumn grove]

I’m struggling a bit with flow, first time trying to use it, already signed up and every time I click on create with flow it only let me watch flow tv, could anyone give me a hand please

[kind moth]

Do you want a task id

With the problem

I kinda lost the one with the many execs

But I have one with a single

Error

[chilly spindle]

Yes please share your task id

[kind moth]

Here's one https://jules.google.com/task/9076784306246564406

[jolly wyvern]

to be honest probably its going to be around $200 as other high quality models out there. but first it needs to go through this beta version as its still quite early, Jules can be struggling sometimes even with basic tasks, sometimes can do good work. if it will be consistently good and devs do good job (which I think they do, they listen to all feedbacks and fix problems) - doubt it would be free, but it can be worth it as it seems like a great tool. idea is good, implementation requires some updates still, but worth waiting for

[gilded scroll]

is it me or there's a lot more "internal errors" with Jules this week? Sometimes it gets stuck running a command and can't even push the changes it has to git (due to git lock)

[spring burrow]

man I hope they don't wall it behind their $250 paywall

that would hurt

[light basalt]

I just tested again and it's still happening.

[mint heart]

A week ago, Jules decided to replace -python-json-logger. Today, it is bringing it back into my code base.

[late shale]

Ultra account, extend auto switch to veo 2. Help !!!!😭😭

[chilly spindle]

Can you please share your task id as well? Team is currently working on this. Also you may try #1374125660859142254 for bug reports 🙂

[chilly spindle]

@vocal shuttle That's an interesting idea. Currently it does not have visibility into chat history. I will take this as a feature request. cc @oak roost

[light basalt]

Ok sure. Here is the task ID 10209558834112918115.

[naive quarry]

jules is probably the best thing to ever touch my repos

[oak roost]

Thanks @wind stream for the suggestion, I will raise a bug internally.

[oak roost]

We're working on this issue, I already raised a bug about this.

[meager ocean]

Someone assist to delete/end this task for me. Its corrupted on my end, booting for ever https://jules.google.com/task/1608248642451099571

[oak roost]

#1374125660859142254 could you add this in here.

[vapid moat]

Oh i get....that a lot

this is a new one

fighting past it with the .md approach and doing it later

[cosmic creek]

Any knows here how to create the viral big foot vlogs

[oak roost]

could you raise this in #1374125660859142254

[oak roost]

Thanks for using jules

[vapid moat]

Its amazing

DEspite the...honestly quite common bugs,it is awesome

I wont be submitting that one as id rather not draw SUPREME attention to my OPSEC setup as that does comprise part of it,but ill submit if it happens again

[woeful pendant]

Can we add the function to sort #1374125660859142254 by "number of likes" (and make it default) so we can avoid users sending duplicates and make it easier for the team to prioritize?

[bold sky]

how can we preview the site that jules made faster

im frightened if i worked with jules for like 2 hour and when i deployed it with vercel or something the site totally crashed

[kind moth]

@bold sky

Ask it to make a commit so you can test

You can do that at any point

It will make a pr at a different than your main branch

So vercel should deploy a preview

[vapid moat]

Ive had very bad luck with asking it to commit

even trying to force it to do so

what is the best way

[kind moth]

it's easy

Always works for me

Let me find you an example

Hm

Stopped loading for some reason lol

It's stuck

@vapid moat

[vapid moat]

ill force it in a bit

and watch it fail

[kind moth]

It shall not

Hahah

But we definitely need a UI button for that

It's a basic task so you don't have to convince AI to do it

Also I wish there was a mode where it automatically commits any changes along the way

To the a branch created for the task

Maybe that would be the correct behaviour so you can test at any point or see if it's stuck

[vapid moat]

...

ok ive tried 1000 diff ways to word it

but that 1 sentence

ill try it later on soemthing more stuck

HOWEVER

how can you commit when you have no.............button

[kind moth]

Yeah, you have to ask it

In the chat

[vapid moat]

i did above

its justn ot visible

its fine it gave me the info i needed in that mess

commit changes or create pr so i can review

lets try it thistime when its strugglin

WHERES THE BUTTON NEO

[vapid moat]

WHERE IS IT

JULES

Just Underpowered Lacking (in Enthsiasium SUBMITTER

"YE I MADE ALL THIS WICKED STUFF BUT NAH U CANT HAVE IT ITS MINE"

[kind moth]

Yeah, you didn't pay

[vapid moat]

i did in training data

[kind moth]

Buy Google AI Pro Max for just 499

[vapid moat]

i also gave it malware

thhank god for sandboxing

nah u see im gonna negotiate

mah training data is that good

watch this and hold my beers

Actually JULES p much did 60% of the work decompiling KEYPLUG

The stager anyway,i got a tip on the payload

so thats google with CN nation state malware code nobody else has

[chilly spindle]

Team is working on this fix https://discord.com/channels/1172568727942860810/1381627547858632725. Can you please share your task id?

[vapid moat]

Will find it in a bit

Ah yes this project/mess isnt sensitive

its just a mess

(ikik it all gets read but id rather not the speciifcs get read by EVERYONE being in csec n all)

Done

[chilly spindle]

@vapid moat Thanks for sharing. I have forwarded it to the team. We will keep you posted.

[vapid moat]

No worries,im not bothered

its an amazing tool

sure its got some kinks

just restart,reprhrase,redo

I keep trying to get it to give me debug info and trying new ways of getting past stuff

but it refuses to give up anything about its environment sadly

[obtuse rock]

Can Jules be configured to use an SSH key via GitHub repository settings or any other method?
In my setup, I'm cloning submodules using SSH due to compatibility with other CI tools. However, Jules doesn't seem to have access to the necessary SSH key, so it fails to clone the submodules and throws an error.
If this topic has already been discussed or documented somewhere, I’d really appreciate it if you could share a reference.

[vapid moat]

Not as far as I can tell it appears to be have a closed ecosystem as in whatever you clone in at the start or You know you can even recursively clone in that's it you can't bring anything in or Send anything out this is a good idea in for malware reasons and other reasons in my opinion even if it is annoying

It would make absolute sense for it to totally lack the ability to connect to anything via SSH for security reasons.

I can't be certain there isn't a Google next to my name unless they want to hire me, which I'm very much open to, but everything else I've seen points towards this extreme level of diligence.

[obtuse rock]

Thanks for the insight! From a design perspective, it sounds like there's little chance that support for SSH key access will be added in the future.
I'll explore other options!

[tall estuary]

I have a project using Deno. But for the life of me can't seem to figure out what to put in the environment setup. Anybody have any tips or a guide on how I can leverage it to get Deno installed? At the moment all my prompts start with install Deno and then the rest of my instructions 🙃

[feral terrace]

guys how tf to get it to push to github? i asked it to make some changes to my code and now it says 'i cannot directly commit to github' - cmon bro.

[kind moth]

...

[feral terrace]

yes?

[kind moth]

^

[vapid moat]

Say commit

1s let me get the exact right thing

Commit changes or create pr so i can review

say those exact words

[small topaz]

Is Jules working right now? Last night, it was able to clone my Rust project, but right now, it's stuck cloning, and it's been almost an hour

i tried multiple tasks but same result

[hybrid thistle]

Working for me but very slow

[feral terrace]

thank you

worked

[polar rock]

Jules has trouble with Flutter :/

[median cloak]

Will Jules have an API? I want to build a linear integration that uses tickets -> fire off tasks 🤔

[chilly spindle]

@median cloak Can you please submit this as a feature request in #1374125660859142254 channel. That way other community members can chime in too. Thank you

[vapid moat]

Will back this with all my soul

WHat little remains

LInk it when youve posted it

[vapid moat]

I have to point out the stellar community engagement and active development going on here btw,idk if i arrived at the right moment but ive never seen such an active dev community listening to its end users
May i ask how large the JULES team is?

Obviously theres a few front facing people but theres going to be good deal more doing backend stuff,but i suspect less than i think as AI is a force multipllier

Either way great work,this would make a good paper in community driven AI agent development at some point

Case study w/e

[vapid moat]

ghost in the machine

will report in a bit

this is ODD

Its like another user is feeding it conflicting commands

[icy marlin]

Hello! Are there any comms around when codecasts will return?

[midnight token]

is this a joke? Whats the point of giving jules whole codebase, when instead of actually using it, it just assumes and halucinates things?

every coding assistant out there when working with a project either indexes it or understand the current scope of given workplace

I have a whole folder dedicaded to brushes and Jules just "oh well, i assume your code is here, I will assume it should work that way, and I will give you a recipe for pierogis and act as this is what you wanted"

[fossil karma]

Simultaneous tasks not clearing, I have no tasks ongoing but get "You have reached your simultaneous task limit of 5 tasks" Anyone know how to clear it?

[kind moth]

Jules is stuck on loading screen

[kind moth]

yes

You have to stop paused tasks

delete them

and tasks at yellow

with query

[fossil karma]

Thanks!

[subtle jetty]

Got a question for the community: Has anyone had experiences with Jules handling code that includes JSDoc comments (after the fact) versus standard comments generated by Jules? I'm torn between implementing "robust jsdoc comments" as I'm unsure adding JSDoc comments to files will actually improve the reliability of AI-generated code enough to outweigh the additional strain JSDoc places on the LLM's contextual memory.

The project currently hasn't reached a point where the use of 100% JSDoc for human use is a contributing factor yet, I'm more focused on improving LLM generated code reliability.

[winter dove]

isnt jules supposed to version test? i asked iit if it could refractor my schemas.py into seperate orginised files and it missed almost all required import

like there was literally 40 different errors after the refractor it broke other files to 😭

[subtle jetty]

I’ve made it a habit to add a task that runs specific tests to check any changes or refactoring I do. When tests already exist, I use them to catch issues early. I’m about to refactor some large (300+ line) files myself, and I’ve been thinking about how to improve my prompts so they also instruct Jules to verify code changes properly (stating explicitly in some way what constitutes an actual acceptable pass condition).

Without a check like that, I know I’d miss a ton of things when doing my refactoring—just like what happened in your case (accept I wouldn't catch them till much later on).

[winter dove]

ah, yea ive been trying to now say specifically to unit test the code. My normal use for it is just updating my documenation for my (ai is really good at doing dirty work like docs) and its worked just fine, its just that one time I asked it to refractor it died lol

[plucky zodiac]

is my code safe?

[oak roost]

could you raise this as feedback here #1374125660859142254

[oak roost]

Ohh...could you raise a bug here #1374125660859142254

[kind moth]

I haven't tested this

But has any of you pushed to Jules repo and told it to pull successfully?

Or should I not try

I'm scared it'll break

[still copper]

What do you mean by Jules repo, sorry? Unless I am missing some context, I create a repo (if new project) on GitHub and then use that in Jules for the task I want to set.

[naive quarry]

Jules has some gaps right now when it comes to project level context, maybe try directing it to look in that folder?

[hoary rain]

what is the generic markdown guidelines when using agents.md? Does anyone know?

[meager knot]

i hope jules is good as code janitor, that is one thing I hate doing

[kind moth]

Yes

But after a single commit

Jules has it's own branch

Can it pull

[still copper]

Yes Jules creates a new branch to push the changes it made. You can create a pull request and merge into your main/master branch on GitHub.

I guess what you mean is if you make changes to this branch that Jules created, will Jules now pull the changes? That is something I have not tried.

[kind moth]

Yea

That's what I mean

[vapid moat]

It will not pull the changes mid-processed, no.

It's very restrictive with the tools it has and how it uses them. For example, if you ask you to do a review of things and it finds critical errors, instead of actually fixing them, it will generate a file, a .md file with the fixed completes in there for you to override it with because it was not designed in that run to actually edit Python files.

If it does not pull everything in the repo in at the start the tools are no longer available So you have to have any other repo you want to pull in in the configuration

Any sort of attempt to connect to anything or download anything externally is pretty much blocked after you initiate a run because of, well, probably security reasons as well. There's like a million other reasons.

[vocal hill]

does this work only ofr Org repos , as I am unable to my personal private repo , with all the right permissions

[rain phoenix]

Have Jules write it itself

The markdown syntax itself shouln't be that important, it doesn't actually process it e.g. to turn *asterisks* into italics.

But it should be legible in a text editor and free from obvious errors

[patent blade]

Any advice for getting the internal errors? Searching doesn't seem to provide much insight into if there's anything I can do. Attempted to clear all the tasks, rebuild teh virtual environment and created new tasks. Just can't seem to get anything to run that's not part of the original virtual environment configuration.

[subtle jetty]

As of today (June 21, 2025), I've encountered an issue with all newly created tasks. Running npm test succeeds the first time, but every subsequent attempt to run npm test fails. This problem does not affect any tasks created before today. This is true for both before and after reverting any github merges made today.

[errant harness]

Is it better to frequently start new tasks ? I figured it would reduce the context size, but spinning up a new vm is taking longer than expected

[wet isle]

Did Jules get better ?

Didn't use Jules since start of may, cause of bugs, like her not posting actual changes.

[subtle jetty]

Been having a lot of issues today. Still can't isolate definitively if they're due to user code, virtual environment related that persist between tasks, or if I have additional lingering issues of jule's leaking it's Environment config generated code (no longer correctly using relative paths)

[wet isle]

Ah im too dumb to understand what that all means 🥲

Jules was my first found AI that could read the whole codebase, without me needing to copy-paste things.

So im not as advanced as you i guess.

[frozen aspen]

Been having a lot if issues today.

[real citrus]

I’ve compared OpenAI’s Codex (on the web) vs. Google Jules. I only made a very general directive to improve some things, but there was additional guidance in the repo. Codex was ridiculous — it picked a few linter-level problems and fixed those. Jules was much more ambitious, and the flow was overall pretty good. Jules managed to fix one issue that I’ve thrown Claude Code CLI, Cursor (with Sonnet, o3 and Gemini 2.5 Pro, in several attempts), but they all failed. Jules prevailed. There are UI annoyances in Jules, and I think the FIRST STEPS are still a bit confusing: tasks and repos/workspaces are listed independently, although a task only runs within one workspace. But OpenAI’s Codex on the web is much worse 🙂 Overall, Jules is a better product that Codex, both UI-wise and functionally!

[silk bobcat]

I have a repo loading issue. I added my one GitHub account to Jules to test it out, and now I want to remove that repo and add another repo from my other GitHub user account. I removed the original repo but when I add the new repo to the new GitHub account, Jules askes me to authenticate, which I do but then it redirects me back from GitHub to Jules but it never loads the newly added repo? Its stuck in some sort of loop, if I select add repo I go through the same process again with the exact same result. Any guidance will be much appreciated?

[real citrus]

MCP support, tools support?

• Question 1 : Does Jules currently have any way to use tools/MCP servers?

• Question 2: Is there a file or folder in the repo equivalent to CLAUDE.md (Claude Code)/ AGENTS.md (OpenAI Codex CLI) / .cursorrules (Cursor) / .github/copilot-instructions.md (VSCode Copilot) that Jules automatically ingests into its context? —— UPDATE: https://jules.google/docs#include-agentsmd-file mentions that Jules uses AGENTS.md , which is the same as OpenAI’s Codex CLI

[tardy night]

we need docs for agents.md guide and use cases

[subtle jetty]

I probably wasn't articulate enough. I've used Windsurf (Codeim), and Github Copilot via VS Code, but am a complete novice when it comes to the Javascript language that my project is based on that I've been using jules for. Saw a bunch of possibilities using Gemini App Builder via aistudio.google.com/apps paired with Jules capabilities when Jules became available.

[subtle jetty]

Oh I believe it. Once my project gets to the point where that becomes more required, I dread thinking all the points of failure that crop up from AI needing to accurately distinguish and reliably use logic to handle code that is interdependent across multiple files.

It makes me shutter thinking about the additional logic required to generate code in a manner that comes with refactoring. A few ways I've thought to address it is not letting files initially go above ~300-500 lines of code, implementing an extreme amount of unit tests so that the AI can deduce what wrong code it's previously generated, or introducing a universal template to the end of prompt in some way that achieves:

#Commenting Guide:
**Comment only if essential:** For complex/non-obvious logic or if code interacts with other files/modules.
**Content:** Explain purpose. If external interactions, name involved files/modules.
**Avoid:** Comments for simple, self-documenting code.

Or something similar when going back and commenting code before refactoring the file. I don't have much experience with LLM's handling refactoring yet and I foresee a bunch of irrelevant redundant comments that may just clog up the contextual memory.

[winter dove]

oh i never even thought of how that could

effect it

from large comments

[chilly spindle]

@real citrus Can you share more on what tools you would like to use? I am happy to take this as a feature request to the team

[real citrus]

Ideally Jules should be able to call remote MCP servers and local MCP servers that get installed in the VM, just like Claude Code. I should be able to specify the config for the MCP servers in the repo. For me this would be useful if Jules is working on a particular repo that's part of a larger project, and there are external sources that Jules may need to consult to understand the situation better.

[charred dune]

How to get updated code from GitHub in jules project?

[zealous talon]

I am also facing the same issue.

[tardy night]

As of now jules uses gemini 2.5 stable and its deep thinking also?

[winter dove]

Facing issues where jules just completly disregaurds imports and does not properly unit test

[halcyon cape]

Is there any way of handling secrets securely in the environment ? This is something that I have been able to do in Codex and would like to do also in Jules. It is just to be able to get Flutter packages that are referenced as git private repos using some Github PAT keys.

[coarse stump]

Any tips on how to get Jules to be better at coding/testing swift code?

[halcyon cape]

Sentry MCP server would be nice to give context on issues, it can speed up solving bugs. Context7 could be nice also for up to date documentation.

[chilly spindle]

Thanks for your report. I have filed a bug for this. You can also posts bugs in #1374125660859142254 channel.

[chilly spindle]

Thanks for your feedback. Can you please post this request on #1374125660859142254 ?

[chilly spindle]

@charred dune Each task runs in a fresh virtual machine where Jules clones your repo. Please see https://jules.google/docs/faq/#how-does-jules-work-under-the-hood . Does this help? Thanks!

[formal tusk]

Alrighty discord jules ppls, I'm back.

ICYMI (or in case I missed this being shared here), we're hosting another live AMA this week: #labs-announcements・📣 message

This time it won't be over chat but over video so that'll be fun. Please be sure to join!

[formal tusk]

what type of docs are you thinking. Getting started docs? FYI docs? FAQ docs?

[chilly spindle]

Can you confirm if you followed this? v

[potent forge]

Flow crashed

[zealous talon]

Yes

[chilly spindle]

@silk bobcat Can you please confirm if you removed jules from your main account

[compact trellis]

what tf?

[snow halo]

Has Jules’s prompting changed recently? It keeps asking me before it creates a plan to tell it the names of files it should be able to check itself. It never used to do this until around a week ago.

For example, it keeps asking me what is the name of my javascript file that handles xyz even though I only have one javascript file and it should know that, and it does.

[frozen aspen]

Has Jules’s prompting changed recently?

[oak roost]

Could you raise this in #1374125660859142254 and share us the task link, we'll create a tracker for this.

[honest flame]

Sounds like you're aiming for a full-stack web app with Jules! 🚀 While it excels at focused tasks, building an entire React+Auth+Firestore app from a single prompt is complex. Try breaking it down into smaller, specific steps. Thanks for your feedback. It is crucial for Jules' growth! 🙏

[rapid pewter]

thank you. does Jules run ci pipeline ? if I put tests into ci would it run it? I'm thinking of trying to put the firebase emulator and yarn dev into environment start up so that Jules can create and run tests

[honest flame]

We're thrilled to hear that Jules outperformed other powerful models like Claude Code CLI, Cursor, and even OpenAI's Codex in tackling a challenging issue you'd presented. That's exactly the kind of complex problem-solving we're aiming for.

We also really appreciate your candid feedback regarding the UI annoyances and the confusion around "FIRST STEPS" with tasks and repos/workspaces. You're right – clarifying that relationship is crucial for a smoother onboarding experience. We're constantly working on improving the user interface and streamlining the initial setup, so your observations here are incredibly valuable.

[honest flame]

Yes, Jules can absolutely run tests. It operates within a secure Cloud VM where it clones your repository and executes commands. So, if you've set up your CI pipeline to trigger tests (e.g., npm test or yarn test), Jules can run those tests as part of its task execution.
Jules has an "Environment setup" section where you can provide commands to prepare the environment. If your tests rely on the Firebase emulator or need the yarn dev server running, configuring these in the setup script would enable Jules to run your tests effectively.
It essentially creates the environment your code needs to function, allowing Jules to then execute your test suite and verify changes. We encourage you to experiment with this setup!

[honest flame]

Starting new tasks frequently would theoretically reduce the context size within any single task, but as you've noticed, the overhead of spinning up a new VM for each task can indeed be longer than expected, a common trade-off when working with autonomous agents like Jules. For iterative work, continuing an existing task is often faster.

[vapid moat]

Actually, it excels at this.

I found it very good at writing function tests, doing them and then modifying the functions based on the tests and iteratively working until it actually, you know, works.

[rapid totem]

So I am having jules write integration tests that require API keys and what not. how can I have jules get the information so jules can verify the written tests with out having keys checked into code.

[fossil karma]

Is it just me or did Jules get smarter in the last 12 hours

[formal tusk]

depends on what you're defining as smarter. What are you seeing?

[fossil karma]

It understands things better, although that might just be me but it's definitely faster and actually commits when it needs to.

[subtle jetty]

In environment settings Id assign the key to a sys variable and have script call that. That's just a rough guess as I'm not sure what the permissions are like for the virtual machine.

[split zodiac]

How jules work on prees the add repo button you select your repo it redirect to the home page and now you can see the repo you selected
What if the frontend is running on localhost and the backend is running on a ngrok domain how frontend will get the access token and the repo is choosed we can see

how things work under the hood

[rose quail]

I really wish there was a download button here.

めちゃめちゃここにダウンロードボタンがほしいよ。

When jules is asking for verification, it is a hassle to create a directory and download files one by one, and it is also a hassle to interrupt the conversation to commit and download the files from the terminal.

julesが検証を頼んできているときにわざわざディレクトリをつくってファイルを一つ一つダウンロードしたりするのは面倒だし、会話を中断してコミットさせてターミナルからダウンロードするのも億劫だと感じる。

[junior brook]

Just so I'm 100% clear: It's considered safe to put secrets (such as API keys) into the "Initial setup" instructions in the repo config? The documentation says don't put secrets in the repo (which, yes, obviously), but doesn't say what the CORRECT way to handle secrets is. 🙂

[elfin breach]

Hi! I’m running into a problem with Maven authentication in Jules.

I’m trying to build a Java project that pulls an artifact from a private JFrog Maven repository. I’ve added the correct credentials to /home/jules/.m2/settings.xml during the initial setup.

However, when Jules runs mvn clean install, it still fails with a 401 Unauthorized while trying to download the artifact from the private repo.

It seems like Jules is ignoring the settings file I create.

Is there a different Maven settings location Jules uses, or does it restrict how settings.xml is loaded?
What’s the correct way to provide Maven credentials in Jules?

[snow halo]

Ive noticed the opposite. It keeps asking obvious questions.

[haughty axle]

Hi! Did anyone have issues when it comes to jules not updating the branches in the repository?

[stray yoke]

Jules generated a 8k line file 😄
https://github.com/JustasMonkev/phpWebShop/pull/1/files#diff-1fee0f3c077c0b1f337edabafcc6f481a2d448f243fba3aa560208609b152ccc

[muted osprey]

I have given permission to github repo but still cannot see it in Jules. Please help me out

[golden stump]

how do i get jules to debug something i am developing in replit?

[formal tusk]

Hey Tom! Great question and unfortunately quite a few steps as Jules is currently only integrated with GitHub. I think you'd have to link your replit --> GH. Once it's in GH you may need to clone it locally to continue development.

But of course all of this is dependent on what you have in replit like are you trying to run it on github or just storing the code there? Eitherway, you'll need to get it into GitHub first and then prompt Jules to debug that repository.

[formal tusk]

Try refreshing. If that doesn't work then we'll need to go through the settings in GH verify that repo has provided access:

GitHub settings --> Applications --> Select "authorised oauth apps' and click on Jules. This was you can review and adjust the rpo access granted. You can choose to allow access to all repos or select specific ones only. Save, refresh, and try!

[formal tusk]

Yesssss love hearing this! Thanks for sharing John.

[formal tusk]

hah! We definitely shipped some changes last week.

just a reminder for tomorrow (cped below):

**## Jules AMA on Discord **Join us for a live AMA with the Jules crew! 🐙 bring your questions, a snack, and hope to see you there! check out the Events tab for more info. <@&1194667985235890277>

• When: <t:1750953600:f>
• Where: #1182419495369855056
• Who: the Jules crew & you 🙂

Bring your snacks and questions! It'll be fun.

[jaunty pendant]

i will be there this going to be so interesting

[cerulean fractal]

If we miss this will it be up on yt or something to watch later?

[formal tusk]

that's a great question. I don't think we typically record these things but defer to @elfin kite . No worries though, we'll be sure to continuously check-in with AMAs with this crew.

[junior brook]

@formal tusk Do you know the answer to the question I asked above? "Is it considered safe to put secrets (such as API keys) into the "Initial setup" instructions in the repo config? The documentation says don't put secrets in the repo (which, yes, obviously), but doesn't say what the CORRECT way to handle secrets is. "

[cerulean fractal]

Thanks 🙏 @elfin kite will it be recorded?

[junior brook]

Unrelated question, but how can I cancel a task? Jules seems to be stuck on one of them.

[formal hatch]

👍